Premium plugin
AI Assistant is not included in the standard Redactor package. It ships as a paid add-on — the Assistant Suite — purchase it on the pricing page (with Redactor, or later from your account), then download the package and enable the assistant plugin.
The AI Assistant turns Redactor into a writing workspace with AI — not a chat box on the side. Authors stay in the document: select a block (or several), open AI in the toolbar, and choose what should happen next. The result comes back as a review, a report, or a ready-to-apply rewrite — without leaving the editor.
For wiring, endpoints, and action maps, see the Assistant plugin reference.
Concept#
The Assistant plugin is the editor UI for AI — menus, review cards, reports, refine, and apply. It does not call OpenAI, Claude, or any other provider directly. On your side you keep an API key and a small backend that receives the editor request, runs your prompts, and returns JSON the plugin can show.
There is no vendor AI subscription and no per-token billing from Imperavi. Requests and tokens are entirely on your infrastructure. You choose the model, rate limits, logging, and which actions exist.
Why the backend is required
Provider API keys must stay private. Putting them in JavaScript would expose them. So Redactor posts to your endpoint (ai.url); that endpoint talks to the model and responds in the shapes the editor expects.
For request/response contracts and copy-paste handlers, see AI Assistant backend examples. For prompt templates per action, see Writing AI Assistant prompts.
Main features#
Suggestions, not silent rewrites. AI proposes an improvement with a short explanation — original and suggestion side by side — so the author can accept, edit, or refine before anything replaces the draft. The same pattern works for translation and other edit actions.
Reports in host tabs. For summary, key points, writing check, and similar actions, your backend turns the model response into HTML and the editor opens it in a separate tab — keep the briefing open while you continue editing.
Inline annotations. Pair with Annotations plugin to highlight words or phrases and offer alternative wording in place.
Generate with options. Ask for additions such as a lead or headings and pick from several variants instead of taking a single forced output.
Flexible UI over your API. The editor supplies the surfaces (review cards, option lists, refine, report tabs). A small CSS framework ships for report HTML so backends can render structured results without inventing one-off markup.
Your backend, your prompts. There is no vendor AI subscription in the plugin. You keep keys and policy on the server, define any actions and prompt templates, and shape responses the way your product needs.
Async jobs when needed. Long documents or slow models can use create/poll jobs — including a separate worker host — so work stays reliable without hitting sync request timeouts.
Edit#
Improve writing, make text shorter or more detailed, simplify language, change tone, or translate. The result is a suggestion to review — not an automatic rewrite of the draft. Preview the change, compare with a diff when you want it, then accept or discard.
A Review card shows the original text, the AI suggestion, and a short explanation of why that change was proposed.
Some actions return several variants at once. Compare them, pick one to apply, or keep writing without using any.
If a suggestion is close but not right, Refine opens a dropdown of adjustments — tighten it, change tone, or ask for another pass without starting over.
The same refine flow can open a textarea so you describe in your own words how the suggestion should change, then send that note back to AI.
Review#
Check writing and similar review actions return a list of concrete issues tied to the document. Instead of a long chat reply, authors walk through items: fix, accept a suggestion, or ignore.
Insights#
Summarize, extract key points, build an outline, or generate meta-oriented copy. These open as report tabs next to the draft — useful for a briefing on a long article without replacing the original text.
Better wording#
Pair the Assistant with Annotations. Better wording marks phrases in place; authors open a highlight and pick an alternative from a callout.
Image generation#
When image AI is configured, generation runs from a form in the editor content area — enter a prompt without switching tools.
After generation, the result appears in the editor. Insert it into the document or try again with a different prompt.
Related guides#
When you wire Assistant into your product, these how-to pages cover prompts, backends, async jobs, and report HTML.
Writing AI Assistant prompts — how to write prompt templates per action so the model returns JSON the editor can apply, review, or show as a report.
AI Assistant backend examples — request/response shapes and copy-paste Node.js and PHP endpoints for OpenAI, Claude, and image generation.
AI Assistant async jobs — create/poll job protocol when sync requests are too slow for long documents or queued models.
AI report HTML framework — CSS classes and markup patterns for report HTML that opens in host tabs.